241 results about "Mixed model" patented technology

The invention discloses a voice synchronous-drive three-dimensional face mouth shape and face posture animation method. A user can input new voice information, and the new voice information can be preprocessed to combine mouth shape animations and face posture animations which are synchronous with voice on the face head of a virtual man. The method specifically comprises two stages. In a training stage, voice visualization modeling can be achieved through a k-nearest neighbor algorithm (KNN) and hidden Markov model (HMM) mixed model. In a combining stage, the user can input new voice information, characteristics of voice signals are extracted, face posture and mouth shape sequence parameters corresponding to the voice signals can be generated through the KNN and HMM mixed model and are processed in a transition mode, and X face open source software is used to combine delicate and abundant three-dimensional face animations. The method has significant theoretical study value and has wide application prospect in the fields of visual communication, virtual meetings, games, entertainments, teaching assistance and the like.

The invention relates to the technical field of face recognition, in particular to a rapid detection method of a face in a high-resolution color image under a complex background. The invention comprises the following steps: building a face skin color mixed model which is composed of two color space restrictions of RGB and YCbCr according to a large amount of acquired skin color sample data to determine the skin color pixel; rapidly skipping non-face regions by adopting a whole skin color pixel ratio to improve the location efficiency of a face candidate region; then using the improved face rapid detection algorithm which is based on forward characteristic selection to realize face preliminary judging of the face candidate region; and finally utilizing a false alarm restraining method which is based on space restriction and geometric restriction to further lower false detection rate and complete face detection. The invention can realize rapid location of a plurality of frontal faces in images at a high detection rate under the condition of low false alarm rate; the good performances of the invention are proved by the results of the test set of Bao open database and a plurality of video frequencies and high resolution color images.

Disclosed is a three-dimensional liver CT image automatically segmenting method based on hyper voxels and the graph cut algorithm. The method comprises analyzing a volume data histogram to adaptively enhance the image contrast; performing primary liver contour segmentation layer by layer through an adaptive threshold and an morphological method, selecting the largest liver segment to compute and extract a liver interest area; selecting seed points on the largest liver segment according to a primary liver contour, modelling foreground and background colors through a Gaussian mixed model; generating hyper voxels on the contrast-enhanced liver interest area through an SLIC (simple linear iterative clustering) algorithm, structuring an undirected weighted graph by taking the hyper voxels as the vertex, and segmenting the undirected weighted graph through the graph cut algorithm; performing postprocessing on segmentation results through a morphological algorithm. The three-dimensional liver CT image automatically segmenting method based on the hyper voxels and the graph cut algorithm can achieve rapid and accurate automatic segmentation of the liver in a three-dimensional abdominal CT image.

The invention discloses a film comment-oriented naming entity recognition method, and belongs to the technical field of entity recognition. The method comprises the following steps of: defining a naming entity type of a name in a film comment; constructing a training data set TrainSet; segmenting a non-marked data set by taking characters as units so as to construct a character matrix, and constructing a mapping dictionary of characters and word vectors by utilizing a skip-gram model in word2vec; generating a naming feature dictionary corresponding to each character; combining word vectors andnaming features of the characters to generate a feature vector matrix for training a name recognition model; constructing a mixed model by utilizing Bi-LSTM and CRF so as to train the name recognition model; and carrying out name recognition on the feature vector matrix through the trained name recognition model, so as to realized film comment name recognition task-oriented name definition, carryout name labeling on character levels during name recognition, obtain name features and combine word vectors obtained via word2vec training with the naming features.

The invention discloses a music similarity detection method based on mixed characteristics and a Gaussian mixed model. According to the basic thought, the method comprises the steps of using a gamma-tone cepstrum coefficient for conducting similarity detection, and using weighting similarities of various characteristics as a final detection result; providing a modulation spectrum characteristic based on a frame shaft, using the characteristic for representing a music long-time characteristic, and using the combination of the long-time characteristic and a short-time characteristic as the input of modeling in the next step; using the Gaussian mixed model for conducting modeling on the music characteristics, firstly, utilizing a dynamic K mean value method for conducting initialization on the model, then, using an expectation-maximization algorithm for conducting model training, obtaining accurate model parameters, and finally using a log-likelihood ratio algorithm for obtaining the similarities between the pieces of music. According to the music similarity detection method, the obtaining of the music characteristics is more sufficient and thorough, the accuracy degree of music recommendation is improved, the characteristic vector dimensionality can be reduced, the information memory content of a music database is reduced, and the accuracy degree of the music recommendation is improved.

The invention discloses a remote sensing object detection method based on a few samples, and the method comprises the steps: firstly carrying out the processing of collected samples, and obtaining anauxiliary data set; secondly carrying out the similarity verification of the auxiliary data set and a target domain; thirdly training DPM (Deformable Part based Model) parameters according to the auxiliary data set, obtaining a target classifier, and carrying out the target detection, wherein the parameters in the model are divided during the training of the DPM parameters, and then the transfer learning is carried out to obtain the target classifier. The method employs a target clustering mode during the training of a source domain data mixed model, can effectively reduce the training errorsand can reduce the training complexity. Moreover, a structural parameter dividing method is employed, so the method can improve the detection performances of an adaptive model.

The invention discloses a low-energy cooperation transmission method in a heterogeneous network. A heterogeneous network system comprises a macro cell and a micro cell. The macro cell comprises a macro base station. The micro cell comprises a micro base station. A cooperation cluster is composed of the macro cell and the micro cell. The low-energy cooperation transmission method in the heterogeneous network includes a first step of obtaining normalization precoding of a Co MP-JP model, a micro base station (MBS) model and a micro base station (m BS) model based on P-ZFBF, a second step of constructing mixed cooperation precoding, wherein the Co MP-JP model, the MBS model and the m BS model are arrayed in a weighted and plural mode to obtain mixed precoding vector quantity of a user i of all base stations, a third step of ensuring data range of compound weighted coefficient amplitude value |Lambada ij| and obtaining circuit power consumption, signal processing power consumption, backbone network power consumption and emission power consumption in the data range and a fourth step of obtaining corresponding total power consumption of seven mixed models in the third step and obtaining a service model which enables the total power consumption to be the minimum, corresponding mixed precoding and the minimum power consumption. The low-energy cooperation transmission method in the heterogeneous network self-adaptively selects a cooperation model with the minimum power consumption, thereby achieving the minimum energy consumption.

The invention discloses a text-related speaker recognition method based on an infinite-state hidden Markov model, which can be used for solving the problem that overfitting or underfitting data is easily generated in the traditional hidden Markov model. The text-related speaker recognition method disclosed by the invention comprises the following steps of: firstly, carrying out preprocessing and feature extraction on a voice signal set for training; then, describing the set for training in a training process by adopting the infinite-state hidden Markov model, wherein the model has an infinite state number before training data arrives and an output probability distribution function corresponding to each state is expressed by using a student's t mixed model; after the training data arrives, calculating to obtain a parameter value in the model and the distribution condition of random variables; and during recognition, calculating a likelihood value related to each trained speaker model on the basis of recognizable voices subjected to the processing and feature extraction, wherein a speaker corresponding to the maximal likelihood value is used as a recognition result. The method disclosed by the invention can be used for effectively improving the recognition accuracy rate of a text-related speaker recognition system, and in addition, the text-related speaker recognition system has better robustness for noises.

The invention discloses a nuclear power plant breakwater overtopping impact simulation method based on a mixed model. The method includes the following steps that at first, a three-dimensional entity finite element model of a fine structure of a breakwater is established; secondly, a breakwater structure rigid body model and a neighboring environment entity model are established; thirdly, initial stress is added to the established three-dimensional mixing finite element model of the nuclear power plant breakwater to simulate the initial operating condition to obtain the initial state of the nuclear power plant breakwater under the action of gravity and seawater pressure; finally, after distribution of initial stress of the nuclear power plant breakwater is obtained, the impact state and the overtopping state of the nuclear power plant breakwater under the action of waves are obtained, and integral simulation of the breakwater is achieved. According to the method, simulation analysis of numerical simulation of the super-large-scale nuclear power plant breakwater is achieved, response rules of an overall model can be contained, meanwhile local structures can be finely analyzed, local dangerous positions are obtained, and reference bases are provided for engineering design of the nuclear power plant breakwater.

The invention, which belongs to the field of speech recognition technology, provides a Sichuan dialect identification method and apparatus, an acoustic model training method and apparatus, and equipment. The acoustic model training method comprises: collecting Sichuan dialect speech data; carrying out feature extraction on the Sichuan dialect speech data to obtain speech features; training the speech features by using a hidden Markov model-Gaussian mixed model, acquiring a classification label corresponding to each frame of speech features and generating to-be-processed speech features with the classification labels; and training the to-be-processed speech features by using a depth delay LSTM model to obtain a target acoustic model. With the acoustic model training method provided by the invention, the time needed by acoustic model training is saved effectively; the training efficiency is improved; and the recognition efficiency and the recognition accuracy are improved.

The invention relates to an embedded, Chinese-English mixed language supporting, embedded application oriented voice recognition method and system for non-specific people. The invention adopts acoustical model trained by mass voice data, acoustical modeling unit set compatible with Chinese and English pronouncing mode, so as to implement Chinese-English mixed voice recognition for non-specific people. According to the invention, a plurality of background models are adopted, a Gauss mixed model (GMM) parameter is obtained by an average adaptive training executed by the background models, then a vector quantization to the difference between the average of the Gauss mixed model and the average of the background models, and the model parameters are compressed. In the recognition stage, rapid Gauss selection, acoustic score pre-calculation, and a simplified GMM model are used, so that the amount of recognition calculation and storage space of the models are greatly reduced, and the voice recognition method and system is applicable on various kinds of embedded application systems.

The invention discloses a polypropylene melt index predicating method based on a multiple priori knowledge mixed model, which fully explores and utilizes priori knowledge of a polypropylene industrial site, and is used for organically integrating various priori knowledge, embedding the priori knowledge into a multilayer perceptron neural network in a non-linear equality constraint form, and optimizing a network weight number by means of a particle swarm optimization algorithm based on an augmented Lagrange multiplier constraint processing mechanism. Based on the multiple priori knowledge neural network model, the multiple priori knowledge neural network model is organically integrated with a polypropylene melt index simplification mechanism model into a harmonic average mixed soft-measuring model. The multiple priori knowledge mixed soft-measuring modeling method has good fitting prediction ability, and is capable of enhancing model extrapolation capacity and realizing good unity of model extrapolation and prediction accuracy of polypropylene melt indexes. Besides, the method is capable of avoiding zero gain and gain inversion and guaranteeing safety in practical polypropylene melt index quality closed-loop control application.

The invention provides an OCT image quality evaluation method based on an improved Resnet and SVR mixed model, comprising the following steps: step 1, pre-processing the original OCT image to obtain the pre-processed OCT image; step 2, construct and training a depth residual network, and adopt that trained depth residual network to extract the deep-level features of the pre-processed OCT image; 3, train an OCT image quality evaluation model by utilizing that deep-level characteristic of the extracted preprocessed OCT image; Given an OCT image to be evaluated, the quality fraction of the OCT image to be evaluated is obtained by using the OCT image quality evaluation model. The invention combines the depth residual network with the quality evaluation task of the OCT image, and establishes anew objective prediction model of the subjective perceived quality of the OCT image.

The invention discloses an optimal method for controlling a mixed model of a first-order reaction continuous stirred tank reactor (CSTR). The method can ensure that the first-order reaction CSTR can rapidly, accurately, stably operate under the following two conditions: a transient process that a system is started and then operates into a certain stable operating point, and a transient process that the system converts into the other stable working point from a certain stable operating point. Compared with the existing first-order reaction CSTR control method, the method provided by the invention has obvious advantages.

The invention discloses a forest above-ground biomass remote sensing estimation general model construction method. Remote sensing data is preprocessed first to obtain a normalized data set with the remote sensing data of Landsat from 1987 to 2011 and 6 periods of measured data as data sources; a sample plot remote sensing factor, the biomass value and a change value are calculated; static and dynamic models of the forest above-ground biomass are constructed through multiple regression, linear simultaneous equations, geographical weighted regression and linear models; and the optimal model is determined as the general model of the forest above-ground biomass through verification, and a parameter form of the model is given. The invention fully exploits the relationship between the change ofthe remote sensing factor and the forest biomass change, therefore, the uncertainty caused by the estimation of single-period remote sensing data is reduced; the forest above-ground biomass model is constructed through a linear mixed model, therefore, the simulation and prediction accuracy is greatly improved; and the constructed general model can be used for real-time and on-site estimates of theforest biomass.

The invention relates to a named entity recognizing method based on a mixed model. The method comprises the following steps: pre-processing; by virtue of a self-adaptive selecting mode, in a hidden Markov model and a conditional random field model, selecting a model with a relatively high F value as a self-adaptive statistic recognizing model, initially recognizing the named entity for a recognized corpus to obtain an initial named entity recognizing result; constructing a basic dictionary formed by a knowledge base and a recognizing rule library; by virtue of the basic dictionary, performing secondary recognition on the initial named entity recognizing result by adopting the self-adaptive static recognizing model, and analyzing the F value of the secondary recognizing result, and updating the basic dictionary; and constructing the mixed model based on the basic dictionary and the self-adaptive statistic recognizing model, recognizing the to-be-recognized corpus to obtain a person name entity, a place name entity and an institute name entity in the to-be-recognized corpus, supplementing the recognizing result into the knowledge base, and updating the basic dictionary for recognition next time. According to the method provided by the invention, the recognizing accuracy and the recognizing recall rate are remarkably improved.

The invention discloses a webpage resource upgrading method of mixed model application. The webpage resource upgrading method is executed in a mobile terminal, an installation package of the mixed model application comprises a webpage resource compression package, and after the mixed model application is installed to the mobile terminal, webpage resources are decompressed into an ROM of the mobile terminal. The upgrading method includes the steps that under preset conditions, a resource updating request is sent to a server side, and resource updating information returned by the server side responding to the resource updating request is received; according to the resource updating information, a webpage resource updating package is downloaded from the server side; when the mobile terminal is started the next time under the mixed mode application, the downloaded webpage resource updating package is decompressed into the ROM of the mobile terminal; the webpage resources are read from the ROM and rendered. The invention further discloses the corresponding mixed model application, the mobile terminal comprising the mixed model application and a webpage resource upgrading system.

The invention discloses a mixed model judgment based crystallizer breakout predication method. The mixed model judgment based crystallizer breakout predication method comprises the following steps of:1) acquiring real-time temperatures of thermocouples inside a crystallizer, and sending the real-time temperatures to a computer terminal; 2) correcting the acquired temperature values and modelling,and converting the temperature values into an DIB image to store; 3) performing discretization on the DIB image, and obtaining crystallizer temperature fields through a thermal imaging system; 4) performing image identification and classification on a temperature field distribution image, and judging whether the temperature of each thermocouple of the crystallizer is normal or not; and 5) comprehensively judging whether the temperature has anomaly or not, and judging the anomaly reasons according to the image identification results and the results of a comprehensive expert system, thereby predicating whether a crystallizer breakout phenomenon appears or not. The mixed model judgment based crystallizer breakout predication method realizes accurate predication and warning of crystallizer breakout during continuous-casting production, and improves quality of continuously cast slabs.

The invention discloses a short-long-term prediction method based on online taxi-hailing travel requirements. The method comprises the following steps of firstly, preprocessing data and segmenting thedata into a training set and a test set; then, dividing an urban road network into grids according to longitude and latitude and finding space-time correlation between areas; next, establishing a mixed model based on CNN+LSTM+XGBoost; and finally, predicting a taxi demand quantity within a short time period (such as 10 minutes) and a taxi demand quantity in a long time period (such as 1 hour, festivals and holidays and rush hours). The method can be used for short-term prediction such as prediction of traffic flow trends with intervals of 10 minutes, and can also be used for long-term prediction; and different periodic changes caused by weekends and festivals and holidays can be considered and different passenger flow rules in the rush hours every day are found out, so that the predictionaccuracy is improved.

The invention discloses single-index prediction of a big data platform based on artificial intelligence in the technical field of artificial intelligence, and the method comprises the following steps:(1) a first layer is a data exchange center, and is responsible for the collection work of data, serves as a basic corpus of big data, and carries out the data mining; (2) the second layer is used for storing data and is responsible for storing the collected data; (3) the third layer is data processing and is responsible for calculating data stored in the HDFS and the HBASE; (4) the fourth layeris used for data display and is responsible for displaying mined data; And (5) the fifth layer is constructed by adopting a mixed model data center based on the main data warehouse and the auxiliary big data processing platform, and the index of the big data platform is pre-warned, so that the fault operation and maintenance efficiency is improved, the loss caused by a major fault is reduced, andthe processing capability of mass data is improved.